gpu: transpose patches faster

[a10y]

Eliminate all of the wasteful allocating and deallocating, replace it with a 3-pass algorithm that reuses the buffers.

Drops the CPU transpose time by a factor of 5-10 in my benchmarks on the other branch

Tested using existing test suite

[codspeed-hq]

Merging this PR will improve performance by 23.62%

⚡ 3 improved benchmarks
✅ 391 untouched benchmarks
⏩ 2052 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	take_map[(0.1, 1.0)]	4.2 ms	3.5 ms	+20.8%
⚡	Simulation	take_map[(0.1, 0.5)]	2.6 ms	2.1 ms	+23.62%
⚡	Simulation	take_map[(0.1, 0.1)]	1,007.5 µs	908.7 µs	+10.88%

---

_{Comparing transpose-patches-fix (c9e1e79) with develop (5d6a3c8)²}

Footnotes

1. 2052 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

2. No successful run was found on develop (761c404) during the generation of this report, so 5d6a3c8 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩